One week. One system. We read what you have built, find what stands between it and production, and hand you a ranked list of the work with what each item costs. If you commission that work from us within 60 days, the fee comes off the first invoice in full.
$5,000 5 working days 1 system 2 engineers
The assessment covers one system end to end - a model you want to deploy, a pipeline that keeps breaking, or a prototype that has to become a product. We are looking for the things that will page someone at 3am, not for style points.
Whether the current model can be reproduced from data and code by an engineer who did not write it. Pinned environments, seeds, data snapshots, and whether the artifact in production matches anything in your repository.
Training and inference paths, orchestration, scheduling and retries. Where state lives, which steps are idempotent, what happens when one dies halfway, and how long you would take to notice.
The validation strategy, the split, and the distance between your offline number and the business outcome. Leakage, temporal ordering, and whether a good score is measuring generalization or measuring your test set. A leaderboard is not generalization, and we have paid to learn that.
How inferences are served and what they cost per request. Batch versus online, model versioning, the rollback path, and what happens to traffic while a bad model is being taken out.
Concept and performance drift, input distribution checks, and whether your alerts fire on things that matter. If the model silently degrades over a quarter, we want to know which graph moves first.
Where features and labels come from, how stale they are by the time the model sees them, and which upstream change silently poisons the whole thing. Label quality gets the same scrutiny as model code.
Both of us work the engagement. You will hear from us on day one and day four whether or not you ask.
A 60-minute kickoff: what the system does, what production means for you, and what is forcing the timeline. We take read access and start reading.
The six areas above, in your code and your infrastructure. We run what we can run, reproduce what we can reproduce, and write down every assumption we are making so you can correct us.
We draft the findings and take them to one of your engineers before they harden. Anything we got wrong about your system gets fixed here, not in the final document.
The written assessment lands, followed by a 90-minute session with your team to argue with it. You leave with a plan you can staff on Monday.
$5,000FIXED
Not an estimate and not a day rate. If the week runs long, that is our problem. Invoiced 50% to hold the week and 50% on delivery, net 14. USD, exclusive of any applicable taxes.
If you commission implementation work from us within 60 days of the report, the entire $5,000 is credited against your first invoice. Take the plan and do the work in-house and you owe us nothing further - that is a perfectly good outcome, and it is why the assessment is priced the way it is.
If the first conversation makes clear that an assessment would not help you, we will say so before you pay for one. We would rather turn down a week than sell you a document you did not need.
Usually not. Read access to the repository, plus a description of how the system is deployed and access to logs or dashboards, covers most of it. Where we do need more, we will ask for the narrowest thing that works and tell you why.
Yes, as long as something exists. A notebook that works, a prototype that has to become a product, a pipeline half-built - all fine.
If the problem is still open and there is no model at all, the assessment is the wrong first purchase. That is research and prototyping: a baseline, an honest evaluation of whether the approach can work, and a prototype worth arguing about. We scope that per problem and quote before starting - usually a two-week feasibility sprint: a first honest baseline, a straight answer on whether the approach can work at all, and a written go/no-go. Same terms as the assessment, and credited the same way if you go ahead.
Both of us, on every engagement. Haseeb leads the production engineering and is accountable for what ships; Mohibullah leads the modeling and evaluation. No work goes to anyone outside the two of us, and there is nobody between you and the people writing the code.
Quoted as fixed-price milestones taken straight from the remediation plan, so you are buying against a list you have already read rather than signing an open-ended retainer. Most of it is pipeline, evaluation and monitoring work; some of it is modeling. First engagements typically run two to six weeks, and nothing starts before you have the number in writing. Your $5,000 comes off the first invoice.
Yes - send yours and we will sign it. We do not make clients use our paper. The only clauses we come back on are blanket non-competes, uncapped liability, and IP assignment drafted broadly enough to swallow tools we already own. If you would rather not draft anything, we can send a short mutual NDA instead.
Backprop LLC is a US company, working across US, UK and Pakistan time zones. In practice that gives a working day of roughly 04:00-17:00 UTC - full cover for the UK and Europe, the whole US Eastern morning, and the US Pacific coast until mid-morning.
On-site weeks in the US are available where they would genuinely help. If you need a standing slot outside our usual window, say so early - we would rather fix it up front than miss you by five hours.
Backprop is a new practice and has none to show. Everything on this site labelled as prior work is employment - at Google, the University of Oxford, The Home Depot, Afiniti and Noon Academy - not Backprop client engagements, and none of those organizations endorse us.
What we can do instead is talk through the technical detail of that work in a call, and point you at what is public: open-source contributions, a granted patent, and published competition results.